A key challenge in molecular simulation is reaching experimentally relevant timescales.  With GPUs, one can now simulate ~50ns/day in all-atom detail on a single GPU.  However, to reach millisecond timescales with kinetic fidelity is still a great challenge, as scaling to use multiple GPUs with traditional molecular dynamics methods is very difficult, due to the relatively slow nature of even the fastest computer networks compared to the computational power of GPUs. Instead, we propose a new approach for simulating long timescale dynamics using Markov State Models (MSMs).  MSMs allow even a modest cluster of 20 GPUs to simulate a millisecond of aggregate dynamics per day and to use that data to quantitatively predict experiment, indicate the degree of robustness of results, and yield novel insights.  I demonstrate this method with applications to all-atom molecular simulations of protein folding and protein conformational change on the millisecond timescale and beyond.

Simulating millisecond dynamics in all-atom detail using a GPU cluster: Protein folding and conformational change of kinases and GPCRs

246th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.